A graph-based data modeling and natural language query system for SAP Order-to-Cash (O2C) data. Explore interconnected business entities visually and query the dataset using natural language.
- Interactive Graph Visualization — Force-directed graph of ~700 nodes and ~1200 edges across 8 SAP entity types
- Natural Language Querying — Ask questions in plain English, get data-backed answers via an NL→SQL→NL pipeline
- Node Exploration — Click any node to inspect metadata, filter by entity type, expand neighborhoods
- Guardrails — Two-layer filtering (regex + LLM) ensures only relevant queries hit the database
Frontend (React + react-force-graph)
│
│ REST API
▼
Backend (FastAPI)
├── Graph Engine (NetworkX) ── in-memory graph traversal
├── Chat Pipeline (Gemini/Ollama) ── NL → SQL → NL
├── Guardrails ── domain relevance filtering
└── SQLite ── 19 normalized tables from SAP O2C data
How the chat works: User question → LLM generates SQL → SQL executes against SQLite → LLM summarizes results in natural language. The LLM never fabricates data — every answer is backed by actual query results.
| Layer | Technology |
|---|---|
| Frontend | React, react-force-graph-2d, Vite |
| Backend | FastAPI, Python |
| Database | SQLite |
| Graph | NetworkX |
| LLM | Google Gemini (cloud) / Ollama (local) |
- Python 3.11+
- Node.js 18+
- Ollama or a Gemini API key (free)
cd backend
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env # then add your GEMINI_API_KEY or configure Ollama
python main.py # runs on http://localhost:8000cd frontend
npm install
npm run dev # runs on http://localhost:5173backend/
main.py API endpoints
database.py SQLite schema and data ingestion
graph.py Graph construction and traversal
llm.py LLM integration (Gemini/Ollama)
guardrails.py Query relevance filtering
frontend/
src/App.jsx Main layout and state
src/components/ GraphView, ChatPanel, NodeDetail